here - Stefan-Marr.de

6.4. Comparing Implementation Size
taken into account, the additional implementation size remains reasonable
and requires only a 35% larger implementation accounting for the added
genericity.
6.4.4. Event-Loop Actors: AmbientTalkST
This section examines the ad hoc as well as the OMOP-based implementation
of event-loop actors. It briefly restates the two implementation strategies (cf.
Sec. 6.2.3 for details), outlines the supported features, and then discusses the
implementation sizes.
Ad hoc Implementation The ad hoc implementation realizes the notion of
safe messaging by implementing AmbientTalk’s far references with stratified
proxies. Thus, actors do not exchange object references directly when message
are sent. Instead, all object references to remote objects are wrapped in a
proxy, i. e., the far reference.
While this implementation achieves a good degree of isolation with low
implementation effort, it does not ensure complete state encapsulation. For
instance, it does neither handle global state, introduced by class variables,
nor potential isolation breaches via primitives. This choice was made to keep
the implementation minimal, accepting the restrictions discussed in Sec. 3.3.2.
OMOP-based Implementation As discussed in Sec. 6.2.3 and as sketched
in Lst. 6.7, the OMOP-based implementation relies on the full feature set provided
by the OMOP.
While it offers the same APIs and features as the ad hoc implementation,
its semantic guarantees are stronger. By utilizing the OMOP, it can provide
the full degree of isolation, i. e., state encapsulation and safe message passing.
Thus, access to global state and use of primitives are properly handled and
do not lead to breaches of isolation between actors. Note that this implementation
does not use proxies to realize far references. Instead, it relies on the
notion of ownership to enforce the desired semantics.
Results and Conclusion As listed in Tab. 6.6, the ad hoc implementation of
AmbientTalkST uses 5 classes with 38 methods. These account for 183 LOC
and have a total of 428 bytecodes. The OMOP-based implementation on the
other hand has 2 classes, which have in total 18 methods with 83 LOC and
190 bytecodes.
167